Quality-of-Service Routing in Ad-Hoc Networks Using OLSR
Ying Ge, Thomas Kunz, Louise Lamont
               CRC       Carleton University         CRC
http://kunz-pc.sce.carleton.ca/
tkunz@sce.carleton.ca

Contributions
Introduce QoS routing into OLSR
Develop three heuristics to enhance OLSR
Evaluate heuristics
Prove the optimality of two of the heuristics in static network case

Related Work in QoS Routing
QoS Route Information
Using traditional best-effort Ad-Hoc routing algorithms
Piggybacking QoS information in control messages
Useful for call admission control
QoS Routing Computation
 Ticket-based probing
CEDAR
Hierarchically structured multi-clustered organizations

Description of OLSR
Selects MPR to cover 2-hop neighbors
Exchanges neighbor/MPR information in Hello message
Generates and relays TC message to broadcast topology information
Reduces control overhead by limiting MPR set
In the graph, B selects C as MPR

OLSR QoS Limitations
Same topology as before, numbers along edges indicate available bandwidth
OLSR: node B will select C as its MPR
So all the other nodes know that they can reach B via C
When D is building its routing table, for destination B, it will select the route D-C-B, whose bottleneck bandwidth is 3, the worst among all the possible routes
Optimal route (i.e., path with maximum bottleneck bandwidth: D-F-B (bottleneck bandwidth of 10)

QoS Modifications
Change MPR selection
Limited by neighbour knowledge: direct neighbours and two-hop neighbours
Assume bandwidth to neighbours and between 1-hop and 2-hop neighbours is known
Change route table calculation, based on TC messages (which contain information about MPR selection)

MPR Selection
OLSR_R1: similar to OLSR (i.e., choose 1-hop neighbours that cover maximum number of 2-hop neighbours), tie-breaker now maximum bandwidth
OLSR_R2: select the best bandwidth neighbors as MPRs until all the 2-hop neighbors are covered.
OLSR_R3: selects the MPRs in a way such that all the 2-hop neighbors have the maximum bottleneck bandwidth path through the MPRs to the current node

Routing Table Calculation
Take bandwidth into account:
Maximum spanning tree (in paper)
Extended Bellman-Ford (BF) shortest path
Same complexity, second algorithm guarantees that we find shortest route as well
Proof that algorithms indeed correctly find maximum bandwidth path
If complete knowledge, proof by contradiction
Can show that TC messages have enough (i.e., the important) information

Evaluation of QoS OLSR
Simulation: generate networks, run OLSR algorithms, compare results against paths calculated by Link-State algorithm (i.e., complete knowledge, all-pair shortest path)
Network area: 1000 M ´ 1000 M
Number of nodes: 100
Transmission range: 100 M, 200 M, 300 M
Bandwidth: assigned randomly
Results are averaged over 100 randomly generated networks

Performance Metrics
Error rate: percentage of routes with non-optimal bandwidth
Average difference: for routes with non-optimal bandwidth, how far off the optimal bandwidth are we
Overhead: the average number of control messages transmitted per node
MPR count: average number of MPRs in the network

Experimental Results

Conclusions
OLSR_R1: simple change, with little additional overhead, but already significant improvement (mostly due to change to routing table calculation)
OLSR_R2 and OLSR_R3 seem to be perfect, paper contains proof that both MPR selection criteria allow to reliably detect optimal path
OLSR_R2 selects fewer MPRs, less overhead, therefore preferred
Transmission range has important impact: higher range means denser network but not necessarily higher overhead or error rate

Future Work
Implement OLSR_R2 in OPNET and study performance in dynamic environment (currently under way)
Trigger TC messages not only when topology changes, but also when significant bandwidth changes
QoS OLSR overhead will compete with data traffic for fixed bandwidth, if overhead too high, best bandwidth route may well have less available bandwidth than non-optimal best-effort OLSR route
Compare QoS routing to QoS signaling in MANET such as dRSVP, INSIGNIA, SWAN
Combine QoS OLSR with SWAN